CEE 218X Final Project
Darren Wong, Alexander Ng
12/03/2021
Analysis of Travel Time in Bay Area and Surrounding Counties
Problem Statement
About 4.6% of Bay Area residents were extreme commuters who travel 90 minutes or more each way to work. As urban housing prices increase, low-income populations are often displaced out of urban areas further away from work locations contributing to longer commute times. This is exacerbated by the lack of quality public transportation in these regions, especially during off-peak hours, such that people who work irregular schedules, usually low-income individuals, have no safe or affordable way to get to work. Squeezed on both ends by housing and commute factors, they lose access to job opportunities, worsening inequality in the Bay Area.
This study seeks to address the following questions:
Do low-income residents in the Bay Area experience longer commute times?
Are there any spatial correlations between measures of affluence (i.e. income, car ownership, home ownership) and commute times?
Are there statistically significant relationships between socio-economic indicators on commuting times?
Do the relationships uncovered contribute to improving policy interventions that reduce commute times or increase access to employment opportunities for low-income residents?
Data Preprocessing
Data on travel time and other socio-economic variables was extracted from Public Use Microdata Survey for 18 counties (9 Bay Area counties and their 9 immediate adjacent counties). We filtered the data to working adults aged 25 and above and removed observations of negative household income and unemployed individuals. Household income was divided into quintiles for subsequent equity analysis while travel time and education were categorised into four buckets.
Equity Analysis of Travel Time across Household Language
The equity analysis shows that English-speaking individuals have a higher tendency to travel less than 30 minutes to work, possibly due to more abundant job opportunities that allow individuals to consider their travel time in their employment choices. On the other hand, individuals who speak Asian or Pacific Island languages tend to work further away from home, with a disproportionate number of them taking between 30 and 90 minutes to travel to work.
Equity Analysis of Travel Time across Income Groups
Using a stacked bar chart, we conducted an equity analysis to compare how travel time differed amongst different income quintiles. Relative to the total proportion of working adults, low-income individuals have shorter commute times as compared to high-income individuals. There appears to be an inverse relationship between household income levels and commute times.
In carrying out our subsequent analysis, we hope to understand whether low-income individuals are:
- Less likely to incur the opportunity costs associated with longer commutes (due to unproductive hours);
- Less likely to have flexible work arrangements (i.e. shift work);
- More likely to reside within close proximity to their work locations;
- Less likely to access job opportunities located farther away from their homes
A line chart was created to better visualize the trends per income quintile. Interestingly, the variance observed between different income groups increases as commute time increases, especially for travel times between 60 and 90 minutes. This suggests that there are diverging push and pull factors influencing an individual’s travel decisions when commute time is longer. For instance, low-income individuals may shoulder other household responsibilities that discourage them from spending more time on travel while higher-income individuals can afford the cost of travel, e.g. fuel.
Geospatial Visualizations
Map of Low-Income Individuals Travelling More than 60 minutes
Map of High-Income Individuals Travelling More than 60 minutes
Map of Individuals who Own a Car Travelling More than 60 minutes
Map of Individuals who Own a Home Travelling More than 60 minutes
Four socio-economic characteristics were spatially plotted: low-income (<20th percentile), high-income (>80th percentile), car ownership and home ownership. From the map, there is a higher percentage of low-income individuals who travel more than 60 minutes in the counties surrounding the Bay Area, such as San Joaquin (7%), Stanislaus (8%) and Merced (9%). Within the Bay Area, the trend is weaker but seen in Solano and Contra Costa counties.
On the other hand, the main regions with long travel times amongst higher-income individuals are concentrated in the Bay Area, especially Contra Costa, Alameda, San Mateo and Santa Clara counties. This is expected as higher-income individuals tend to work in heavier-populated Bay Area locations and are expected to encounter worse traffic conditions.
Car ownership is higher the further from denser urban regions in the Bay Area as observed in Brentwood (Contra Costa) , San Jose, Tracy (San Joaquin). This suggests that there may be a lack of commute alternatives in the area, such as public transport. These observations are very similar amongst individuals who own a home, which is consistent with the characteristics of these areas as suburban neighborhoods with more working adults financially and physically able to manage long commutes to work.
Regression
##
## Call:
## lm(formula = JWMNP ~ HINCP + AGEP + TEN + JWTRNS + JWRIP + WKHP +
## WAGP + HHL, data = study_pums_reg)
##
## Residuals:
## Min 1Q Median 3Q Max
## -70.607 -14.928 -3.720 6.773 131.446
##
## Coefficients: (1 not defined because of singularities)
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -7.301e-01 8.816e-01 -0.828 0.407537
## HINCP -4.898e-06 8.842e-07 -5.540 3.04e-08 ***
## AGEP -2.618e-02 8.591e-03 -3.047 0.002310 **
## TEN2 -7.879e-01 3.185e-01 -2.474 0.013369 *
## TEN3 -3.078e+00 2.449e-01 -12.568 < 2e-16 ***
## TEN4 -3.633e+00 1.032e+00 -3.519 0.000434 ***
## JWTRNS1 6.978e+01 1.058e+01 6.597 4.24e-11 ***
## JWTRNS10 1.541e+01 9.121e-01 16.895 < 2e-16 ***
## JWTRNS11 -1.311e-02 7.534e-01 -0.017 0.986121
## JWTRNS12 3.775e+01 1.248e+00 30.246 < 2e-16 ***
## JWTRNS2 5.139e+01 8.406e-01 61.134 < 2e-16 ***
## JWTRNS3 5.028e+01 8.585e-01 58.567 < 2e-16 ***
## JWTRNS4 7.080e+01 1.083e+00 65.387 < 2e-16 ***
## JWTRNS5 4.468e+01 1.538e+00 29.055 < 2e-16 ***
## JWTRNS6 6.508e+01 2.058e+00 31.624 < 2e-16 ***
## JWTRNS7 2.495e+01 2.063e+00 12.092 < 2e-16 ***
## JWTRNS8 2.662e+01 1.758e+00 15.142 < 2e-16 ***
## JWTRNS9 2.263e+01 1.082e+00 20.904 < 2e-16 ***
## JWRIP1 -4.000e+01 1.056e+01 -3.789 0.000152 ***
## JWRIP10 -2.116e+01 1.133e+01 -1.868 0.061803 .
## JWRIP2 -3.590e+01 1.057e+01 -3.398 0.000679 ***
## JWRIP3 -3.274e+01 1.059e+01 -3.092 0.001988 **
## JWRIP4 -3.472e+01 1.066e+01 -3.256 0.001132 **
## JWRIP5 -2.874e+01 1.077e+01 -2.670 0.007597 **
## JWRIP6 -2.820e+01 1.146e+01 -2.461 0.013871 *
## JWRIP7 -3.132e+01 1.140e+01 -2.747 0.006025 **
## JWRIP8 -1.423e+01 1.257e+01 -1.133 0.257408
## JWRIP9 NA NA NA NA
## WKHP 9.838e-02 9.297e-03 10.581 < 2e-16 ***
## WAGP 1.855e-05 1.487e-06 12.481 < 2e-16 ***
## HHLEnglish only -1.101e+00 2.786e-01 -3.951 7.79e-05 ***
## HHLOther Indo-European languages -7.694e-01 4.226e-01 -1.821 0.068688 .
## HHLOther language -6.082e-01 9.331e-01 -0.652 0.514519
## HHLSpanish -1.638e-02 3.488e-01 -0.047 0.962542
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 23.61 on 51575 degrees of freedom
## Multiple R-squared: 0.2295, Adjusted R-squared: 0.229
## F-statistic: 480.1 on 32 and 51575 DF, p-value: < 2.2e-16A multiple regression analysis was conducted with the socio-economic indicators identified:
HINCP: Household income
AGEP: age
TEN: tenure type (1- owned with mortgage/loan, 2- owned free and clear, 3- rented, 4- occupied without payment of rent)
JWTRNS: means of transport to work (1- car/truck/van, 2- bus, 3- subway/elevated rail, 4- long-distance train/commuter train, 5- light-rail street car or trolley, 6- ferry boat, 7- taxi cab, 8- motorcycle, 9- bicycle, 10- walk, 11- work from home, 12- others)
JWRIP: vehicle occupancy (number of persons carpooling)
WKHP: number of hours worked each week
WAGP: personal annual wages
HHL: household language.
From the results, household income is negatively correlated with travel time, reflecting the financial considerations of fuel or other familial responsibilities. However, higher earning individuals appear to be more willing to accept the cost of longer commutes. The older the individual is, the shorter the commute time, possibly due to their financial ability to afford faster commute options or the personal choice to work closer to home.
As anticipated, individuals who elect to commute by car have shorter travel times while individuals who travel by public transport have longer travel times. Individuals who travel by taxi, motorcycle or bicycles have the longest travel times, suggesting that these forms of commute are less reliable overall. In addition, carpooling is correlated with increased commute times as there may be multiple, indirect work trips within a specific carpool commute.
Individuals who speak English tend to have a shorter commute time. This suggests that these individuals have better access to job opportunities nearer to home.
Finally, the hours worked per week appears to have a positive correlation with commute times. Longer hours are associated with longer commute times, which could be attributed to irregular hours like shift work that restricts an individual’s commute choices.
The distribution of residuals is mostly normal. This indicates that the chosen model does not over or under fit the data.
After plotting the mean of the residuals, this suggests that the model more accurately predicts commute times within the central Bay Area and is less accurate in the regions farther away. Negative residuals in the northern part of the study area show that the model over-predicts travel time, so there could be fewer commuters travelling to work into the Bay Area or the area may have better transport infrastructure that reduces congestion. On the other hand, positive residuals in the central part of the study area show that the mode under-predicts travel time so there is worse congestion there, which is consistent with the higher percentage of car owners travelling more than 30 minutes living in the area.
Multiple Regression Model with Replicate Weights
| rowname | base | stderr |
|---|---|---|
| (Intercept) | -1.0714897 | 0.7101350 |
| AGEP | -0.0294540 | 0.0095221 |
| HHLEnglish only | -0.5882721 | 0.3187073 |
| HHLOther Indo-European languages | -0.7417386 | 0.4169019 |
| HHLOther language | -0.2505486 | 1.0863349 |
| HHLSpanish | 0.3886781 | 0.4722299 |
| HINCP | -0.0000046 | 0.0000009 |
| JWRIP1 | -41.3035930 | 6.7702350 |
| JWRIP10 | -20.9176157 | 9.3517478 |
| JWRIP2 | -36.2275969 | 6.7576210 |
| JWRIP3 | -33.8230963 | 6.7598281 |
| JWRIP4 | -34.8951660 | 7.3588064 |
| JWRIP5 | -13.3747527 | 11.2997796 |
| JWRIP6 | -25.7311152 | 10.1085975 |
| JWRIP7 | -30.3211560 | 8.1346706 |
| JWRIP8 | -5.2440820 | 15.6483064 |
| JWRIP9 | NA | NA |
| JWTRNS1 | 71.0648659 | 6.7089525 |
| JWTRNS10 | 15.9404332 | 0.4820686 |
| JWTRNS11 | 0.0513053 | 0.1352890 |
| JWTRNS12 | 35.0337056 | 1.5514538 |
| JWTRNS2 | 50.0892336 | 0.9736497 |
| JWTRNS3 | 50.2981985 | 0.7415479 |
| JWTRNS4 | 71.2496166 | 1.6596368 |
| JWTRNS5 | 44.8188652 | 2.0417181 |
| JWTRNS6 | 64.2472956 | 2.0222839 |
| JWTRNS7 | 23.6641030 | 1.5325784 |
| JWTRNS8 | 24.5114237 | 1.1980927 |
| JWTRNS9 | 21.9064180 | 0.7724925 |
| TEN2 | -1.0438100 | 0.3952120 |
| TEN3 | -2.9893608 | 0.3331046 |
| TEN4 | -4.0307639 | 1.4261379 |
| WAGP | 0.0000211 | 0.0000018 |
| WKHP | 0.0959285 | 0.0109175 |
Accounting for replicate weights (WGTP1:WGTP80), standard deviation generally decreased across all variables while model coefficients remained largely similar.
Final Considerations
Some parameters could be considered for subsequent analysis:
American Community Survey with more granular geographical sub-units does not have travel time data for all desired counties despite expanding time period range (missing data sets for 7 counties).
Using LODES data for further Origin-Destination analysis would complement this analysis. However, the chosen dataset was too computationally complex due to large file size.
Incorporating other intersecting social indicators such as education level, race and environmental considerations may provide a more holistic comparison.